named crash many times a day - Memory leak ?

[Jim Reid]

>>>>> "Sebek" == Sebek Pavel <Pavel.Sebek at i.cz> writes:

    Sebek> Hi, one year ago it told here about "named crash many times
    Sebek> a day - Memory leak ? " I have the same problem (but I have
    Sebek> BIND 8.2.2-P5 and Tru64 4.0E) with crashing.

Hmm. I don't think 8.2.2P5 was out a year ago, so if someone was
complaining about frequent crashes then, I doubt if their problem is
related to yours. [Check the mailing list archives at www.isc.org to
be sure.] It's also highly unlikely that there's a memory leak in
8.2.2P5. If there was, we would have almost certainly heard about it
by now and seen a patch for it. FYI I believe that f.root-servers.net
runs 8.2.2P5 on some flavour of Tru64 Unix, so if this platform did
have a memory leak, someone just might have noticed it long before
today. :-)

    Sebek> I'm running on Alpha with Tru64 V4.0E.  All memory options
    Sebek> in my config file are set to default and my OS is quite
    Sebek> well tuned. I also never observed delays, DNS is working
    Sebek> well apart it is crashing many times a day.

Maybe those default VM resource limits are now too low? Try increasing
them. Pay close attention to the OS-enforced limits. Did you configure
the name server to ask the OS for as mucn VM as possible? ie: Is there
an entry like
	datasize unlimited;
in the options{} statement for named.conf?

If your name server just picks up the system defaults, these could be
too low for a memory-hungry name server. [Or maybe the system
administrator has mistakenly reduced those defaults?] What else has
changed on this system recently? Have you suddenly added lots of
clients or an enormous number of resource records to some zone? Has
the OS complained about VM or RAM problems?

It is unusual for a well configured name server to run out of memory.
It's even more unusual for that to be caused by a memory leak. There
will probably some local OS configuration issue that accounts for the
problem. A memory leak would tend to suggest there was a generic bug
in the code that would affect everybody. But it seems to be just your
name server that's running out of memory. That would tend to suggest
there's a local problem with your server and not some memory leak that
affects a substantial number of the world's name servers.